Projection Explanation

I just wanted to add a brief explanation to my seat projection below. The way I'm doing this involves looking at the 2004, 2006 and 2008 results at the riding level and comparing those results to the national and provincial results on a party by party basis. This allows me, given access to some decent polling, to produce an estimate of what would happen if those numbers prove accurate on a riding by riding basis. As I said, I have compiled the data for all 308 ridings. However, I just want to show what the data produces for a couple of ridings to give you an idea of what I'm looking at when I make my predictions.

First a riding where there is little debate over what is going to happen: the Liberal stronghold of St. Paul's in downtown Toronto. Given the National and Provincial data available at threehundredeight (and thanks to the folks behind it for saving me the work), my spreadsheet tells me that the expected result in St. Paul's if the election were to happen with those national/provincial results is:

Liberals (presumably the good doctor): 54.03%
Tories: 21.16%
NDP: 11.06%
Greens: 10.27%

I am well aware this does not add up to 100% but neither do most polls including the aggregates from which I'm working. This result, I think, would be in keeping with any objective assessment of the situation on the ground in St. Paul's. Second, let's look at one of the ridings that raised an eyebrow in the comments: Churchill. Once again based on the national and provincial available (I have a formula to determine what a regional "Prairie" number means in Manitoba for each party), the results are predicted to be as follows:

Liberals: 34.95%
Tories: 29.96%
NDP: 22.79%
Greens: 5.04%

Now, I don't necessarily believe the Liberals will win by 5 percentage points in Churchill. However, if they Liberals really are at 23% in the Prairies (up 4 points from 2008) as the polls suggest, the votes are much more likely to turn up in Churchill than they are Battlefords-Lloydminister where I predict just 12.48% for the unlucky Liberal candidate. That's what my spreadsheet is telling me. So, I'm not saying that you can bank on four Liberal seats in Manitoba, I'm saying that it is a distinct possibility if the polls remain the same. Quick side note, two of the other Liberal seats projected in Manitoba have a slimmer margin than Churchill so it could easily just be Anita Neville coming back for the Grits. Finally let's look at a riding that I think is actually poorly served by my models. In 2008, the results in Edmonton-Strathcona were as follows:

NDP 42.6%
Tories: 41.6%
Liberals: 9.1%
Greens: 6.4%

That result represented a 10 point jump in NDP popularity in the riding since 2006 and an amazing 33 point jump since 2004. With the NDP down marginally in popularity both provincially and nationally, my model anticipates the following for a future election:

Tories: 38.51%
NDP: 33.18%
Liberals: 16.59%
Greens: 8.04%

The problem is that the 2004 result for the NDP (around 9% of the vote) ways down their chances of holding on to the riding in my formula. My model places the majority of the weight on the 2008 election, but 2004 does figure into the calculation because it provides more data points which hopefully makes the predictions more accurate. In this case, because of the massive shift in support, it may be undervaluing the popularity of Linda Duncan in Edmonton-Strathcona. For most ridings, the voter intentions are relatively stable and these kinds of problems don't occur. I stand by my numbers but I understand that they aren't and frankly cannot be perfect. Only time will tell how well this model holds up when faced with reality. I think it is probably a more detailed model at the local level than is generally used to do seat projections and that's why I've decided to share it. Much like DemocraticSpace's riding projections, these should be taken with a grain of salt.